FastMCP 3.0

FastMCP 3.0

21/01/2026
Development & Infrastructure
Welcome to FastMCP 3.0! - FastMCP
The fast, Pythonic way to build MCP servers and clients.
gofastmcp.com

Table of Contents

FastMCP 3.0

FastMCP 3.0 is a framework for building smarter AI context apps, not just tool servers. It allows you to pull tools and data from anywhere, reshape them, control access, track state, and run long-running tasks—with hot reload, versioning, and observability built in for production use.

Features

  • Pythonic Decorators: Use @mcp.tool to instantly turn standard Python functions into MCP-compliant tools with automatic JSON schema generation.
  • Advanced State Management: Built-in support for session-scoped state and “Resources,” allowing agents to remember context across multiple interactions.
  • Prefect Horizon Integration: One-click deployment to Prefect Horizon, a managed platform that handles auth, scaling, and governance for your MCP servers.
  • Observability & Tracing: Native OpenTelemetry (OTEL) support to debug exactly what your agent sent and received.
  • Hot Reload: Develop faster with a server that auto-restarts when you modify your code, similar to modern web frameworks.

How It Works

  1. Install: pip install fastmcp
  2. Decorate: Add @mcp.tool to any Python function (e.g., a database query or API wrapper).
  3. Serve: Run fastmcp dev server.py. The framework automatically generates the MCP protocol endpoints.
  4. Connect: Point your MCP client (like Claude Desktop, Cursor, or a custom agent) to the local server url to start using the tools immediately.

Use Cases

  • Local Agent Workflows: Giving your local LLM (via Cursor or Claude) safe access to your local file system, database, or internal CLI tools.
  • Enterprise “Skills” Library: Building a centralized repository of approved tools (e.g., “Reset User Password,” “Query Sales Data”) that can be shared across your team’s AI assistants.
  • Long-Running Tasks: Executing complex background jobs (via the “Docket” feature) that an agent can trigger and check back on later.

Pros & Cons

  • Pros: Drastically simplifies the verbose MCP protocol; built by the trusted Prefect team (workflow experts); excellent developer experience (DX) with type hints and hot reloading.
  • Cons: Python-only (Javascript/Go developers must look elsewhere); requires familiarity with the Model Context Protocol (MCP) concepts; “Horizon” hosting is a paid service (though self-hosting is free).

Pricing

  • Free (Open Source): The FastMCP framework is MIT licensed and free to use locally or self-host.
  • Prefect Horizon: Managed hosting platform with enterprise features (SSO, audit logs) likely follows a usage-based or seat-based pricing model (contact for details).

How Does It Compare?

  • vs. Official MCP SDK (Python): The official SDK is low-level and verbose, requiring manual schema definitions. FastMCP is “batteries-included,” handling the boilerplate automatically via decorators, much like FastAPI vs. raw Starlette.
  • vs. LangChain: LangChain is a massive orchestration framework for building agents. FastMCP is a lightweight protocol implementation for serving tools to those agents. You often use them together: LangChain agents consuming FastMCP tools.
  • vs. Mcp-rs / TypeScript SDK: If you prefer Rust or Node.js, you should use mcp-rs or the official TypeScript SDK. FastMCP is specifically optimized for Python developers who want a “FastAPI-like” experience.

Final Thoughts

FastMCP 3.0 is the “FastAPI for AI Agents.” Just as FastAPI made building web APIs in Python a joy, FastMCP makes building Agent Tools intuitive and production-ready. For Python developers looking to expose their internal systems to LLMs like Claude or OpenAI, this is currently the gold standard for developer experience. It abstracts away the complexity of the underlying protocol, letting you focus on the logic of your tools.

Welcome to FastMCP 3.0! - FastMCP
The fast, Pythonic way to build MCP servers and clients.
gofastmcp.com